Skip to main content
Announcements
See why Qlik was named a Leader in the 2024 Gartner® Magic Quadrant™ for Data Integration Tools for the ninth year in a row: Get the report

Calculating rolling n-period totals, averages or other aggregations

cancel
Showing results for 
Search instead for 
Did you mean: 
Gysbert_Wassenaar

Calculating rolling n-period totals, averages or other aggregations

Last Update:

Nov 30, 2022 9:45:07 AM

Updated By:

Sonja_Bauernfeind

Created date:

May 4, 2013 3:53:31 PM

Attachments

A question that gets asked regularly is how to calculate a rolling measure over a period of N-months (or weeks or days). For example a 12-month rolling total or a 4-week rolling average. There are several ways to do this. But these approaches have some limitations that need to be kept in mind. I'll try to explain these approaches and some of their limitations.

  • Accumulation
  • RangeXXX functions
  • Set analysis
  • AsOf tables

First let's load some sample data. The SalesData table below will contain sales amount values for twelve consecutive months.

SalesData:
load * inline [
Month, Amount
1,6
2,4
3,7
4,3
5,4
6,9
7,5
8,7
9,8
10,6
11,9
12,7
];

This is a very simple table with little data, but this enough for demonstration purposes.

Once this data is loaded it's possible to create a straight table chart object to display the amount per month and a running total. As expected Month is used as dimension. The expression sum(Amount) will display the amount per month. Now let's add an expression to calculate a running total over three month periods.

This can be done in two ways. The first uses the Accumulation option for expressions. The same expression sum(Amount) is used, but now the Accumulation option is set to 3 Steps Back:

rolling_2.png

The second option uses the rangesum function. That expression looks like this:

rangesum(above(sum(Amount),0,3))

This sums the Amount value on current row and on the previous two rows. The resulting straight table looks like this:

rolling_1.png

This looks good. The rolling 3 months amount is calculated correctly. But what happens if a selection of months is made?

rolling_3.png

The rolling 3 month amount for month 4 is now 3 instead of 14. This is because month 1,2 and 3 are no longer included in the calculation for the rolling 3 month total.

The accumulation option has another issue. It only works when only one dimension is used in the straight table. The rangesum expression can be modified so it can calculate across dimension borders, but the accumulation option can't. The modified rangesum expression adds the total keyword to the above() function:

rangesum(above(total sum(Amount),0,3))

This goes some way to doing what we want, but the issue of displaying the wrong rolling 3 month amount for month 4 isn't solved yet. Contrary to what I first thought there is a solution for this, as Henric pointed out to me in the comments below. By combining the rangesum with the aggr function it's possible to calculate the correct rolling 3 month amounts for each month. The expression needed for that looks like this:

sum(aggr(rangesum(above(total sum({<Month=>}Amount),0,3)),Month))

Read Elif's blog post Accumulative Sums for a more complete explanation.

How about set analysis expressions?

This expression should calculate the sum of amount for the three month period:

sum({<Month={'>=$(=only(Month)-2)<=$(=only(Month))'}>}Amount)


But notice the only() function. This requires that only one month value is selected. After selecting month 4 the result looks like this:

rolling_4.png

This shows the selected month, but also the two previous months. And the values are not accumulated.

Ok, but what about the max function instead of only?

sum({<Month={'>=$(=max(Month)-2)<=$(=max(Month))'}>}Amount)


That gives a different result, but still not what we're looking for:

rolling_7.png

Now only the last three months are shown and again the values are not accumulated.

The 'problem' is that the set is calculated once for the entire chart, not per row. This means that it's not possible here to use Month both as a dimension and in the set modifier in the expression.

There's still an option left to discuss: AsOf tables.

The AsOf table links a period with all the periods in the rolling period. In this example months are used, but it can be applied to any type of period like hours, days or weeks.

For the three month periods needed for a rolling 3 month total this means a month should be linked to itself, the previous month and the month before the previous month. The only exceptions are the first month, which is itself the rolling 3 month period, and the second month that together with the first month is its rolling 3 month period. There are no months before the first month so the first two months cannot run over 3 months.

The AsOf table needed for the rolling 3 month calculations looks like this:

rolling_5.png

This table can be created like this:

AsOfMonth:
load
Month as Month_AsOf,
Month + 1 - IterNo() as Month
Resident SalesData
while IterNo() <= 3;

right join load Month Resident SalesData;

What this does is create three records for every month using the while statement. But that also creates three records for month 1 and 2. This would create a month 0 and a month -1. The right join is used to remove those incorrect month values.

Now that the AsOfMonth table is created the Month_AsOf field can be used instead of the Month field in the straight table. The expression for the straigh table is simply sum(Amount).

rolling_6.png

The straight table now shows the correct rolling 3 month total for month 4.

This can be expanded a little so not only the rolling 3 month can be shown, but also the amount for the month itself. To achieve this the AsOf table is modified by adding a field to label the type of period. And records are added to the table so each Month_AsOf value is linked to the matching Month value:

AsOfMonth:
load 'Current' as Type,
Month as Month_AsOf,
Month as Month
Resident SalesData;

Concatenate (AsOfMonth)
load 'Rolling 3' as Type,
Month as Month_AsOf,
Month + 1 - IterNo() as Month
Resident SalesData
while IterNo() <= 3;

right join load Month Resident SalesData;

There are now two types of periods available: Current and Rolling 3. Additional period types can be added for example for Rolling 6, Rolling 12 month and Year-to-Date periods. You can find examples of these types in the attached AsOf Table Examples.qvw document.

The period type can be used in the chart expressions to calculate the amount for the wanted period:

Current amount: sum({<Type={'Current'}>}Amount)

Rolling 3 month amount: sum({<Type={'Rolling 3'}>}Amount)

Concluding, there are two solutions that do what we want:

1. The rangesum-aggr combination

2. The AsOf table

The first has the advantage that no changes to the data model are needed. It's also possible to dynamically change the period to aggregate over by using a variable instead of a hardcoded number of periods. A disadvantage is that that it's a somewhat complicated expression that also comes with a performance cost.

The AsOf needs changes in the data model to create the AsOf table and fill it with the necessary records. The advantage is that it likely performs better on large data sets. It's also quite versatile since you can add several sets of records to meet different scenario's. The expressions you end up with in the charts are also less complicated. That said, it will likely take you some time to fully understand the AsOf table concept and realize all the places where you can put it to good use.

In the end you'll have to decide for yourself which solution is appropriate in your situation. With regards to the performance of one or the other solution, you will simply have to test to discover if the performance is acceptable. But of course such testing is already part of your development process, right?

I'd like to thank John Witherspoon for introducing me to the AsOf tables concept and Henric for pointing out the solution using the rangesum function in combination with the aggr function.

Tags (2)
Labels (2)
Comments
Gysbert_Wassenaar

For 13 months instead of 3 use while IterNo() <= 13; instead of while IterNo() <= 3;

Not applicable

Hi Gysbert, if I change it to what you have above, would the following addmonths function still work because the calendar year I am using splits the entire year into 13 periods. For example, Period 1 would be from January 5, 2014 to February 1, 2014, Period 2 would be from February 2, 2014 to March 1, 2014, etc.

(addmonths(Period,1-iterno()),'YYYYMM') as Period1

Gysbert_Wassenaar

If your periods are not months then you cannot use any function that works with months. So using addmonths would give you wrong results.

Not applicable

Henric, your addition is exactly what I was looking for. Thanks to you and Gysbert.

Not applicable

Hello Everyone, this idea behind the calculation of rolling n-periods is simply great. But I think this valid for only one year of data. I can be wrong, but when I run this logic for say 3 years of data, 2012, 2013 and 2014 the logic gets implemented to only 2012. The data source is an excel file and 2012's data are loaded first.

I have two questions. A. Could you please explain, how this method is to be used for all the years of data I have.

B. Say, I am in middle of the year, and I want to see 13 periods before my current period, will it be possible to achieve such view where data encompasses between two years. Say, I am in mid of 2014, will it chart be able to show details till mid 2013.

Thanks and any help will be really appreciated.

Gysbert_Wassenaar

You're right that using only the month field you would run into year boundaries. Therefore you need another kind of field that can cross year borders, e.g. a year-month field with values like 2012-12, 2013-01, 2013-02 etc.

See AsOf Table Examples.qvw for some examples that span more years.

stevedark
Partner Ambassador/MVP
Partner Ambassador/MVP

Thanks Gysbert for sharing.

I have done a blog post that gives three examples of using extra dimensions in the data model in order to aggregate values.  There is also a downloadable QlikView with this blog post to illustrate the examples.  You can read it here: http://bit.ly/1nkhVi4

Not applicable

I wonder if everybody have just ignored my post suggesting similar solution with an extra dimension table or was it not visible? Or may be my English is too terse ...

http://community.qlik.com/thread/23746

Not applicable

Hi

I am New to Qv.

Very Useful Post and i had a similar problem of calculating rolling Totals although for next 3 months.. ie. 31/07/2014  should calculate total of Aug, Sept and Oct. I tried using Interval Match by loading an inline data and it is giving me perfect results.

for eg,

Int:

LOAD * INLINE [

    Start, Stop, Next90   

    01-07-2014, 30-09-2014,'31-06-2014'

    01-08-2014, 31-10-2014,'31-07-2014'

];

I want to use 'Month_AsOf' but for Next 3 months. Can u help?


Not applicable

Thanks found the solution,

Just manipulated the script a little for rolling next 3 Months,

AsOfMonth:

Load

Month as Month_AsOf,

Month + IterNo() as Month

Resident MasterCal

while IterNo() <= 3;

Version history
Last update:
‎2022-11-30 09:45 AM
Updated by: